The Three-Dimensional Decomposition of Volatility Memory

Battle Count: 78

Highly relevant for quantitative trading. The framework directly addresses volatility forecasting, VaR/ES calibration, and tail-risk timing—core components of trading strategy development and risk management. The state-dependent memory decomposition provides interpretable signals (regime gates, tempo compression, fractional memory) that can inform dynamic position sizing, stop-loss calibration, and regime-aware trading. The empirical results show G-Clock excels at ES timing on equities and G-FIGARCH dominates variance forecasting on FX, offering actionable model selection guidance. However, the daily-frequency focus and lack of intraday validation limit direct applicability to high-frequency strategies.

High complexity. The framework involves: (1) three distinct gated recursions with different functional forms (logistic, fractional kernel, exponential time-deformation); (2) pairwise and tri-gate combinations requiring careful parameter partitioning; (3) QMLE with forward-accumulation gradients for each gate; (4) Whittle regularization for fractional components; (5) rolling-window estimation with truncation schedules; (6) reparameterization for constraint enforcement; (7) numerical stability measures (gradient clipping, winsorization). The theoretical proofs (ergodicity, identification, asymptotic normality) are rigorous but the estimation code requires custom extensions beyond standard rugarch/fGarch packages.

The paper specifies estimation in R using rugarch and fGarch libraries with custom routines for gates. Rolling window (Tw=1500), FIGARCH truncation (K<=200), and Gaussian QMLE are documented. However, no GitHub repository or code is provided. Data sources (Bloomberg, Yahoo Finance) are named but not distributed. Parameter reparameterization details and implementation notes (gradient clipping, winsorization) are described but full code is absent.